Page 283 - Data Science Algorithms in a Week

264 Oloruntomi Joledo, Edgar Gutierrez and Hatim Bukhari

Based on business model of Lending Club, these four variables were employed in our
framework to determine which borrowers are screened into or permitted to do
transactions on the platform. The NN also has two hidden layers with 5 and 3 neurons
respectively. Finally, the output layer has two neurons called Output 1 and Output 2 that
fire up any value between 0 and 1. Thus, if Output 1 is larger than Output 2 then it is
considered an acceptance, otherwise it is a rejection.
Taking that into account a test with the entire dataset is run and the resulting error is
0.1118. That means that about 11.1% instances of the training values are misclassified.
To improve the capacity of the NN to represent the information and get better results, the
structure of the NN is changed by adding more layers and varying the number of neurons
per layer.
To improve the capacity of the NN to represent the information and get better results,
the structure of the NN was changed by adding more layers and varying the number of
neurons per layer. The new results for a sample of the accepted data obtained an average
training error of 0.009570 and a target error of 0.0100.

Figure 2. Network structure of the neural network.

Agent-Based Simulation and Validation

The individual behaviors of consumers are modeled in the ABS subsystem. The
simulation begins by declaring and initializing all variables. Probabilities are assigned to
the different agent variables based on their corresponding distributions. The loan lifetime
is defined by parameter Term. The requested Amount, FICO, DTI and Credit History are
stochastic characteristic of a borrower.